An artificial intelligence big data algorithm-based management optimization method and system

Abstract

The application relates to the technical field of manufacturing warehouse management, and discloses a management optimization method and system based on an artificial intelligence big data algorithm. The method is realized by edge intelligent decision, multi-source data fusion correction, and a local control right transfer mechanism in a communication exception, and is realized by an edge computing device arranged in a manufacturing warehouse environment (an integrated operation area composed of a production and processing assembly line, an automated stereoscopic warehouse, and a material conveying line), a hardware acquisition module (a plurality of sensors such as photoelectric reflection and ultrasonic distance) on an industrial control bus, an upstream material conveying mechanism driver, and a central data storage server, so that real-time intelligent monitoring and adjustment of material flow are realized, operation efficiency, stability, and risk resistance are significantly improved, and a more robust and intelligent management optimization scheme is provided for a discrete manufacturing and automated warehouse integrated enterprise.

Description

Technical Field

[0001] This application relates to the field of manufacturing and warehousing management technology, and more specifically, to a management optimization method and system based on artificial intelligence big data algorithms. Background Technology

[0002] At the end of the month or during batch settlement, departments such as finance, planning, and warehousing simultaneously launch a large number of multi-table join queries on the central database. The connection pool and computing power are filled with long-cycle report tasks, triggering the high-load protection mechanism of the system, which limits the rate of ordinary data channels and queues new transactions. As a result, the photoelectric sensors at the automated warehouse's inlet cannot write the "material ready" signal to the central warehouse in time after detecting that the pallet has arrived, and the inventory view still shows that there are spare storage spaces. Based on this, the scheduling program mistakenly judges that the backend has the ability to receive materials and continues to send finished product warehousing instructions to the production line. The drive mechanism then continuously pushes the material boxes onto the inlet conveyor line. However, at this time, the stacker crane at the inlet has not yet completed the previous inlet warehousing closed loop. The physical connection area exceeds the carrying capacity due to the continuous influx of materials, triggering the end overflow protection switch, cutting off the control circuit of the main motor for inlet, causing the entire transmission line to shut down in a chain reaction. At the same time, the stacker crane does not receive the material handshake signal and the execution program is suspended, ultimately causing the entire automated warehouse to fall into a physical standstill where it is in a dilemma of inlet and outlet. Summary of the Invention

[0003] This application discloses a management optimization method and system based on artificial intelligence big data algorithms, which aims to solve the problem that in the manufacturing and warehousing environment, under high concurrency access pressure, the data exchange interface of the central database experiences communication timeouts and response delays, resulting in a disconnect between the warehouse location occupancy view and physical reality, which in turn leads to material accumulation, conveyor line shutdowns, and the overall shutdown of the automated warehouse.

[0004] The technical solution of this application is as follows:

[0005] Firstly, this application discloses a management optimization method based on artificial intelligence big data algorithms. This method is executed by an edge computing device deployed in a manufacturing and warehousing environment, which includes production lines, automated storage and retrieval systems (AS / RS), and material handling lines. The edge computing device communicates with hardware acquisition modules within a buffer area, drivers of upstream material handling mechanisms, and a central data storage server via an industrial control bus. The method includes the following steps:

[0006] The photoelectric through-beam sensor and ultrasonic distance sensor in the hardware acquisition module collect multi-source sensor data of material entering and leaving the buffer area, and upload the multi-source sensor data to the edge computing device through the industrial control bus.

[0007] On the hardware computing module of the edge computing device, a multi-source data fusion algorithm is deployed to fuse multi-source sensor data, identify material entry and exit events, and determine the initial quantity of materials in the buffer area.

[0008] When the difference between the initial quantity and the previous valid count result exceeds the preset allowable deviation range, the edge computing device triggers the image recognition module to perform visual detection on the buffer area. Based on the visual detection result, the initial quantity is corrected to obtain the corrected value of the material quantity.

[0009] On the hardware computing module of the edge computing device, a load-bearing judgment machine learning model trained based on historical multi-source sensor data is run. Based on the comparison between the correction value of the material quantity and multiple preset thresholds, the load-bearing status level of the buffer area is determined. According to the load-bearing status level, a graded adjustment command is sent to the driver of the upstream material conveying mechanism through the industrial control bus. The graded adjustment includes at least reducing the speed of material entering the buffer area or stopping the material from entering the buffer area.

[0010] When the correction value of the material quantity in the buffer area meets the preset recovery conditions, the control restriction on the upstream material conveying mechanism is automatically released through the edge computing device, and material conveying is restored.

[0011] The communication module of the edge computing device detects the communication status of the central data storage server. When the communication status is abnormal, the device locally blocks the central scheduling command from driving the upstream material conveying mechanism and transfers control of the upstream material conveying mechanism to the load status level judgment result.

[0012] Secondly, this application also discloses a management optimization system based on artificial intelligence big data algorithms, applied in a manufacturing and warehousing environment, which includes production lines, automated storage and retrieval systems (AS / RS), and material conveying lines. The system includes:

[0013] The edge computing device communicates with the hardware acquisition module, the driver of the upstream material conveying mechanism, and the central data storage server in the buffer area via an industrial control bus. The edge computing device includes a hardware computing power module and a communication module.

[0014] The hardware acquisition module, including a photoelectric through-beam sensor, an ultrasonic distance sensor, and an image recognition module, is used to acquire multi-source sensor data and visual inspection images of materials entering and exiting the buffer area, and upload them to the edge computing device via an industrial control bus.

[0015] The signal acquisition unit, deployed on the hardware computing module of the edge computing device, is used to fuse multi-source sensor data, identify material entry and exit events, and determine the initial quantity of materials in the buffer area. When the difference between the initial quantity and the previous valid count result exceeds the preset allowable deviation range, the image recognition module is triggered to perform visual detection. Based on the result of the visual detection, the initial quantity is corrected to obtain the corrected value of the material quantity.

[0016] The load-bearing judgment unit is deployed on the hardware computing module of the edge computing device. It is used to run a load-bearing judgment machine learning model trained based on historical multi-source sensor data. Based on the comparison between the correction value of the material quantity and multiple preset thresholds, it judges the load-bearing status level of the buffer area. According to the load-bearing status level, it sends a graded adjustment command to the driver of the upstream material conveying mechanism through the industrial control bus. The graded adjustment includes at least reducing the speed of material entering the buffer area or stopping the material from entering the buffer area.

[0017] The control release unit, deployed on the hardware computing module of the edge computing device, is used to automatically release the control restrictions on the upstream material conveying mechanism and restore material conveying when the correction value of the material quantity in the buffer area meets the preset recovery conditions.

[0018] The communication detection unit, deployed on the hardware computing module of the edge computing device, is used to detect the communication status of the central data storage server through the communication module. When the communication status is abnormal, it locally prevents the central scheduling command from driving the upstream material conveying mechanism and hands over the control of the upstream material conveying mechanism to the bearing status level judgment result.

[0019] This application deploys an edge computing device in a manufacturing and warehousing environment, communicating with hardware acquisition modules, upstream conveyor drives, and a central data storage server. This enables real-time acquisition and fusion processing of multi-source sensor data from material entry and exit buffer areas. Photoelectric and ultrasonic sensors identify material entry and exit events and maintain initial quantities. When necessary, an image recognition module is triggered for visual correction, addressing the problems of environmental interference and accumulated errors in traditional counting. Furthermore, a load-bearing judgment machine learning model trained on historical multi-source data intelligently determines the load-bearing status level based on the relationship between correction values ​​and thresholds, issuing graded adjustment commands to the drives to reduce speed or stop, thus preventing excessive material accumulation and the resulting conveyor line and automated warehouse shutdowns. A mechanism is also designed to automatically release control restrictions when recovery conditions are met, and to intercept central commands locally and autonomously control the load-bearing status level when communication with the central database fails, significantly improving the system's robustness under complex operating conditions and communication failures. Overall, through edge intelligence, multi-source fusion, machine learning judgment, and localized control, refined and intelligent management of material conveying is achieved, ensuring production continuity, stability, and safety. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating a management optimization method based on artificial intelligence big data algorithms provided in an embodiment of the present invention;

[0021] Figure 2 This is a flowchart of a method for issuing graded adjustment instructions provided in an embodiment of the present invention;

[0022] Figure 3 This is a schematic diagram of the structure of a management optimization system based on artificial intelligence big data algorithms provided in an embodiment of the present invention. Detailed Implementation

[0023] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0024] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0025] In integrated discrete manufacturing and automated warehousing enterprises, resources across the entire plant are centrally managed by a unified management system. The central database, upon which these systems rely, can reach its limit during month-end settlements, batch accounting, or cross-departmental report aggregation due to a large number of complex multi-table join queries initiated concurrently by multiple departments. This triggers a high-load protection mechanism, limiting the rate of ordinary data channels and queuing new transactions, causing continuous timeouts for underlying lightweight status update requests. When photoelectric sensors at the automated warehouse's inbound entrance detect a pallet's arrival, they fail to promptly write a "material ready" signal to the central warehouse, resulting in the inventory view still showing available storage space. The scheduling program, misinterpreting this, assumes the backend has receiving capacity and continuously sends finished product inbound instructions to the production line, causing the production line drive mechanism to continuously push the material boxes onto the inbound conveyor track. However, at this time, the stacker crane at the inlet has not yet completed the previous inlet closed loop. The material transfer area exceeds the carrying capacity limit due to the continuous influx of materials, triggering the end anti-overflow travel protection switch, cutting off the main motor control circuit of the inlet, and the entire transmission line stops in a chain reaction. At the same time, the stacker crane is suspended from executing the program because it has not received the material handshake signal, ultimately causing the entire automated warehouse to fall into a physical standstill where it is in a dilemma of entry and exit.

[0026] In this regard, refer to Figure 1 , Figure 1 This is a flowchart illustrating a management optimization method based on artificial intelligence and big data algorithms provided in an embodiment of the present invention. The method is executed by an edge computing device deployed in a manufacturing and warehousing environment, which includes a production line, an automated warehouse, and a material conveying line. The edge computing device is communicatively connected to a hardware acquisition module in a buffer area, a driver for an upstream material conveying mechanism, and a central data storage server via an industrial control bus. The method includes:

[0027] S11, the photoelectric beam sensor and ultrasonic distance sensor in the hardware acquisition module are used to collect multi-source sensor data of material entering and leaving the buffer area, and the multi-source sensor data is uploaded to the edge computing device through the industrial control bus;

[0028] S12, on the hardware computing module of the edge computing device, the deployed multi-source data fusion algorithm is run to fuse the multi-source sensor data, identify material entry and exit events, and determine the initial quantity of materials in the buffer area;

[0029] S13, when the difference between the initial quantity and the previous valid count result exceeds the preset allowable deviation range, the edge computing device triggers the image recognition module to perform visual detection on the buffer area, and the initial quantity is corrected according to the result of the visual detection to obtain the corrected value of the material quantity;

[0030] S14, on the hardware computing module of the edge computing device, a load-bearing judgment machine learning model trained based on historical multi-source sensor data is run. Based on the comparison relationship between the correction value of the material quantity and multiple preset thresholds, the load-bearing status level of the buffer area is determined. According to the load-bearing status level, a graded adjustment command is sent to the driver of the upstream material conveying mechanism through the industrial control bus. The graded adjustment includes at least reducing the speed of material entering the buffer area or stopping the material from entering the buffer area.

[0031] S15, when the correction value of the material quantity in the buffer area meets the preset recovery conditions, the control restriction on the upstream material conveying mechanism is automatically released by the edge computing device to restore material conveying;

[0032] S16, through the communication module of the edge computing device, the communication status of the central data storage server is detected. When the communication status is abnormal, the central scheduling command is blocked from driving the upstream material conveying mechanism locally, and the control of the upstream material conveying mechanism is handed over to the bearing status level judgment result.

[0033] The management optimization method of this application is achieved through the following steps:

[0034] First, multi-source sensor data on material entering and exiting the buffer area is collected using photoelectric through-beam sensors and ultrasonic distance sensors in the hardware acquisition module. The photoelectric through-beam sensor detects whether material has passed through a specific location, generating a pulse signal; the ultrasonic distance sensor measures the distance between the material and the sensor, providing information on the material's stacking height or position. This multi-source sensor data is uploaded to the edge computing device in real time via an industrial control bus. For example, when material enters the buffer area, the photoelectric through-beam sensor triggers a signal, while the ultrasonic distance sensor continuously outputs distance values. This data is received by the edge computing device and used for subsequent processing.

[0035] Secondly, on the hardware computing module of the edge computing device, a multi-source data fusion algorithm is deployed to fuse the multi-source sensor data, identify material entry and exit events, and determine the initial quantity of material within the buffer area. The multi-source data fusion algorithm can employ techniques such as Kalman filtering and Bayesian fusion to comprehensively analyze information from different sensors, improving data accuracy. For example, by analyzing the pulse signals from a photoelectric beam sensor and the distance value sequences from an ultrasonic distance sensor, the direction and quantity of material entry and exit can be accurately determined. Determining the initial quantity is the foundation for subsequent load-bearing status assessments.

[0036] Furthermore, when the difference between the initial quantity and the previous valid count exceeds a preset allowable deviation range, the edge computing device triggers the image recognition module to perform visual inspection of the buffer area. Based on the visual inspection result, the initial quantity is corrected to obtain a corrected value for the material quantity. This correction mechanism is designed to address potential misjudgments or omissions by the sensor. For example, when there is a significant deviation between the sensor count and the actual situation, the image recognition module can be activated to take a picture of the buffer area using a camera. Through image processing and analysis technology, the quantity and location of the material can be accurately identified, thereby correcting the initial quantity.

[0037] Next, on the hardware computing module of the edge computing device, a load-bearing judgment machine learning model trained based on historical multi-source sensor data is run. This model determines the load-bearing status level of the buffer zone based on the comparison between the corrected value of the material quantity and multiple preset thresholds. The load-bearing judgment machine learning model can employ algorithms such as support vector machines and neural networks. By learning the relationship between features such as material quantity, discharge rate, and upstream / downstream cycle time in historical data and the load-bearing status level, it achieves intelligent judgment of the current load-bearing status. For example, when the corrected value of the material quantity exceeds a certain threshold, the system may determine it as "early warning load"; when it exceeds a higher threshold, it is determined as "full load lockout". Based on the load-bearing status level, graded adjustment commands are issued to the driver of the upstream material conveying mechanism via the industrial control bus. These graded adjustments include at least reducing the speed at which material enters the buffer zone or stopping material entry into the buffer zone. For example, when it is determined to be "early warning load", a command can be issued to reduce the conveying speed; when it is determined to be "full load lockout", a command can be issued to stop conveying to avoid material accumulation.

[0038] Furthermore, when the corrected quantity of material within the buffer zone meets the preset recovery conditions, the edge computing device automatically releases the control restrictions on the upstream material conveying mechanism, resuming material conveying. Recovery conditions may include the material quantity falling below a safe threshold or receiving a discharge completion signal. For example, when material in the buffer zone is removed and the quantity drops to a safe level, the edge computing device automatically releases the restrictions on the upstream conveying mechanism, allowing it to resume normal operation.

[0039] Finally, the communication module of the edge computing device detects the communication status of the central data storage server. When the communication status is abnormal, the edge device locally prevents the central scheduling commands from driving the upstream material conveying mechanism, and transfers control of the upstream material conveying mechanism to the load status level judgment result. This mechanism is designed to ensure the autonomous decision-making capability of the edge side in the event of a central server communication failure, avoiding problems caused by delays or errors in central commands. For example, when the central server network is interrupted, the edge computing device will ignore the scheduling commands from the center and instead control the upstream conveying mechanism based on the locally judged load status level, thereby ensuring the continuity and safety of production.

[0040] This embodiment is applied to the precision bearing warehousing process of an automotive parts manufacturing company. The production line produces approximately 12,000 bearings daily, and the automated storage and retrieval system (AS / RS) employs a six-story, twelve-aisle structure. A buffer zone, six meters long and 1.2 meters wide, is set up between the production line and the AS / RS entrance. The buffer zone uses a roller conveyor structure, with the connecting platform's drive motor speed-regulated by a frequency converter supporting the Profinet protocol. The upstream material conveying mechanism is a belt conveyor approximately 15 meters long. A pair of photoelectric through-beam sensors (response time 2 milliseconds) are installed at both the buffer zone's entrance and exit. Three ultrasonic distance sensors (range 80-800 mm, sampling frequency 10 Hz, accuracy ±2 mm) are installed at equal intervals above the buffer zone. An image acquisition device integrating a visible light camera and a polarized light camera (both 6 megapixels, frame rate 30 frames per second) is installed 2.5 meters directly above the buffer zone, at a distance from the material conveying surface. The edge computing device uses an Advantech MIC-770V3 industrial computer, equipped with an Intel Core i7-12700 processor and an NVIDIA Jetson AGX Orin hardware computing module (200 TOPS). It communicates with various hardware acquisition modules, upstream belt conveyor drivers, and frequency converter drivers through the Profinet industrial control bus, and communicates with the central data storage server through Gigabit Ethernet.

[0041] The first step involves the edge computing device acquiring multi-source sensor data on material entering and exiting the buffer area via photoelectric beam sensors and ultrasonic distance sensors in the hardware acquisition module. The photoelectric beam sensors output a falling-edge pulse signal each time a piece of material passes through, while the three ultrasonic distance sensors upload a sequence of distance values ​​at a frequency of 10 Hz. All data is uploaded to the edge computing device via the Profinet industrial control bus and stored in a circular buffer with a five-second time window.

[0042] The second step involves the edge computing device running a multi-source data fusion algorithm on its hardware computing module. This algorithm employs a finite state machine-based decision engine, with states including idle, entry detection, exit detection, interference judgment, abnormal alarm, and visual detection trigger. When the inlet photoelectric sensor captures a falling edge pulse, the state machine switches to the entry detection state, initiating a material passage timer with a maximum duration of 1.5 seconds to continuously monitor the distance value sequence from the ultrasonic distance sensor at the inlet. If the distance value sequence is less than the preset distance threshold of 300 millimeters three consecutive times (a total of 300 milliseconds), it is considered a valid material entry event, the material quantity in the buffer area is incremented, and the state machine switches back to the idle state. If the timer expires but the distance value sequence does not change continuously, it is determined to be environmental interference, the material quantity remains unchanged, the pulse is marked as a false signal, and the state machine switches back to the idle state. When the outlet photoelectric sensor captures a falling edge pulse, similar logic is used, and the material quantity is decremented. When only the distance value sequence changes continuously but the photoelectric sensor does not emit a pulse, it is determined to be an abnormal material stacking height or a sensor blind zone, the material quantity remains unchanged, and an abnormal alarm signal is output. The initial calculated value of the material quantity is denoted as N_t, and it is compared with the previous valid count result N_(t-1). The allowable deviation range is set to positive and negative two pieces according to the material's cycle time.

[0043] Third, when the difference between the initial quantity and the previous valid count exceeds a preset allowable deviation range, the edge computing device triggers the image recognition module for visual detection. The image acquisition device switches to full-frame mode, and the visible light camera and polarized light camera simultaneously acquire a pair of registered images. The edge computing device performs contrast-limited adaptive histogram equalization enhancement processing on the regular image, and performs Stokes vector method polarization calculation on the polarized light image to obtain the polarization degree image of the material surface. The two are fused to generate an enhanced image of the material. The enhanced image is input into the MaskR-CNN material recognition machine learning model accelerated by TensorRT to obtain the material contour and region segmentation results. The edge computing device performs connected component analysis on the mask, removing noise regions with an area smaller than 20 square millimeters and instances with a confidence level lower than 0.7. The remaining number of instances is the material quantity N_v obtained by visual detection. N_v is used as a correction value to overwrite the original initial quantity N_t to obtain the corrected value of the material quantity.

[0044] The fourth step involves the edge computing device running a load-bearing judgment machine learning model trained on historical multi-source sensor data on its hardware computing module (using XGBoost gradient boosting tree, maximum tree depth of eight, learning rate of 0.05, average AUC of 0.98 or higher for five-fold cross-validation, and single inference time of less than five milliseconds). The corrected material quantity N_v, the real-time operating frequency f_up of the upstream belt conveyor, and the real-time current I_dock of the connecting platform drive motor are input into the model to obtain the load-bearing status level of the buffer area. Three preset thresholds correspond one-to-one with three levels: a normal load upper limit threshold of twenty pieces corresponds to normal load, a warning load upper limit threshold of twenty-eight pieces corresponds to warning load, and a full-load lockout threshold of thirty pieces corresponds to full-load lockout. Based on the load status level, the edge computing device sends tiered adjustment commands to the belt conveyor driver via the Profinet industrial control bus: Under normal load, a rated speed operation command is issued, causing the belt conveyor to run at its rated speed of 1.5 meters per second, with the frequency converter driver outputting a frequency of 50 Hz; under warning load, a low-speed operation command is issued, reducing the belt speed to 0.6 meters per second, and the frequency converter driver outputting a frequency of 20 Hz; under full load lockout, a power-off command is issued, and the edge computing device outputs a safety relay disconnect signal via Profinet, cutting off the main power supply circuit of the belt conveyor driver and stopping material from entering the buffer zone. Tiered adjustment includes at least two forms: reducing the speed at which material enters the buffer zone, or stopping material from entering the buffer zone altogether. When the model output is inconsistent with the threshold-based rule judgment result, the edge computing device adopts the model output and records the model decision flag in the local log.

[0045] Fifth, when the corrected value of the material quantity in the buffer area meets the preset recovery conditions, the edge computing device automatically releases the control restriction on the upstream material conveying mechanism and resumes material conveying. Specifically, when the corrected value of the material quantity in the buffer area drops from the critical difference threshold of 28 pieces to below the warning difference threshold of 21 pieces due to receiving the discharge detection signal, the edge computing device initiates the no-load self-check verification process of the docking station: a low-speed jogging control signal lasting two seconds is sent to the frequency converter driver controlling the drive motor of the docking station via Profinet, with a jogging frequency of 5 Hz; the actual output current value fed back by the frequency converter driver is collected at a sampling interval of one millisecond through the analog input port, resulting in a current sampling sequence of approximately two thousand sampling points, and the average operating current during the jogging test is calculated. In this embodiment, the typical value is approximately 1.8. The average operating current is compared with the preset no-load upper limit current threshold (1.56 amperes, determined by multiplying the measured no-load current of 1.2 amperes by a safety factor of 1.3). If the average operating current is greater than the no-load upper limit current threshold, the discharge detection signal is determined to be a false signal, and the docking station is not actually cleared. The power supply circuit of the upstream material conveying mechanism is kept disconnected, and a jamming alarm signal is output. If the average operating current is less than or equal to the no-load upper limit current threshold, the docking station is determined to have sufficient material receiving capacity. The power supply to the upstream material conveying mechanism is restored through Profinet, allowing the production line to resume the material pushing and unloading operation.

[0046] Step six: The edge computing device detects the communication status of the central data storage server through its communication module. Specifically, the communication module initiates a lightweight write transaction test request to the central data storage server every five seconds (writing a record containing a millisecond-level timestamp, a random sequence number, and a sixteen-byte payload to a random primary key row of a dedicated heartbeat table), measuring the round-trip processing time of the write transaction (normally ten to thirty milliseconds). When the round-trip time for three consecutive heartbeat cycles exceeds one hundred milliseconds, or any single round-trip time exceeds five hundred milliseconds, it is determined that the core transaction processing engine is under excessively high processing load, and the communication status is abnormal. When communication is abnormal, the edge computing device sets the scheduling command channel to a restricted verification reception state, locally preventing the central scheduling command from driving the upstream material conveying mechanism, and handing over control of the upstream material conveying mechanism to the bearing status level judgment result. Under restricted verification reception, each scheduled instruction undergoes dual verification using both timestamp and version number: the start timestamp T_start of communication blockage and the end timestamp T_end of communication recovery are recorded, and the instruction generation timestamp T_gen is obtained. If T_gen is between T_start and T_end, it is determined to be a backlogged instruction and its execution is directly blocked. The instruction view version number V_instr on which the instruction is based is obtained and compared with the locally maintained physical verification version number V_local. If V_instr is less than V_local, it is determined to be an expired instruction and its execution is directly blocked. Newly generated valid instructions that pass both verifications are allowed to proceed under controlled conditions, and V_local is updated to V_instr after execution.

[0047] The actual operation data of this embodiment at the auto parts company shows that material spill events have decreased from an average of four times per month to zero, the number of false shutdowns caused by false signals from the docking station has decreased from an average of six times per month to less than once per month, and the number of mis-scheduling events caused by communication blockage has decreased from an average of three times per month to zero. The warehousing efficiency of the automated warehouse has increased by about 15%, which not only meets the production cycle requirements, but also significantly improves the robustness and safety of buffer area management.

[0048] In summary, this application significantly improves the operational efficiency, stability, and resilience of the manufacturing and warehousing environment through edge intelligent decision-making, multi-source data fusion and correction, and a local control transfer mechanism in case of communication anomalies, providing a more robust and intelligent management optimization solution for enterprises integrating discrete manufacturing and automated warehousing.

[0049] Specifically, the aforementioned load-bearing status levels include at least normal load-bearing, early warning load-bearing, and full load locking; the multiple preset thresholds correspond one-to-one with the load-bearing status levels;

[0050] refer to Figure 2 , Figure 2 This is a flowchart of a method for issuing graded adjustment instructions provided in an embodiment of the present invention. S14 includes:

[0051] S141, when the load-bearing state is normal load-bearing, the edge computing device sends a rated speed running command to the driver of the upstream material conveying mechanism through the industrial control bus;

[0052] S142, when the load-bearing state is a warning load-bearing state, the edge computing device sends a low-speed operation command to the driver of the upstream material conveying mechanism through the industrial control bus to reduce the operating speed to a preset low-speed operation level;

[0053] S143, when the load state is full load lockout, the edge computing device sends a power-off command to the driver of the upstream material conveying mechanism through the industrial control bus to disconnect the power supply to the upstream material conveying mechanism and stop the material from entering the buffer area.

[0054] The load-bearing status levels are further subdivided into at least three states: normal load, warning load, and full-load lockout. Normal load means the material quantity within the buffer zone is within an ideal or safe range, allowing for smooth reception of upstream materials. Warning load means the material quantity within the buffer zone is approaching its maximum load-bearing capacity, requiring preventative measures to avoid overload. Full-load lockout means the material quantity within the buffer zone has reached or exceeded its maximum load-bearing capacity, necessitating immediate cessation of upstream material transport to prevent overflow or blockage. Multiple preset thresholds are used to define these load-bearing status levels; for example, normal load occurs when the material quantity is below a first threshold; warning load occurs when it is between the first and second thresholds; and full-load lockout occurs when it is above the second threshold.

[0055] In practical applications, when the load-bearing state of the buffer zone is determined to be normal, the edge computing device sends a rated speed operating command to the driver of the upstream material conveying mechanism via the industrial control bus. The rated speed operating command aims to ensure that the upstream material conveying mechanism operates at its designed or preset normal operating speed, ensuring that materials can enter the buffer zone at a stable rate that matches the production cycle, thereby maintaining the continuity of the production line.

[0056] Furthermore, when the load condition is determined to be a warning load, the edge computing device sends a low-speed operation command to the driver of the upstream material conveying mechanism via the industrial control bus. The purpose of the low-speed operation command is to reduce the operating speed of the upstream material conveying mechanism to a preset low-speed operation level, thereby slowing down the speed at which materials enter the buffer zone. This provides downstream processing time in the buffer zone, preventing materials from accumulating rapidly in a short period of time, thus effectively mitigating the potential overload risk.

[0057] In a preferred embodiment, when the load condition is determined to be full-load lockout, the edge computing device sends a power-off command to the driver of the upstream material conveying mechanism via the industrial control bus. The power-off command aims to immediately disconnect the power supply to the upstream material conveying mechanism, causing it to completely stop operating and thus completely stopping material from entering the buffer zone. This measure is an emergency response to severe overload in the buffer zone, and can minimize serious consequences such as material spillage, equipment damage, or production line shutdown.

[0058] The proposed solution refines the load-bearing status levels of the buffer zone into normal load, early warning load, and full load lockout, and matches each level with differentiated graded adjustment instructions, thereby achieving refined management of the material conveying process.

[0059] In some preferred embodiments, a specific example is given below. Assume a buffer zone in a manufacturing and warehousing environment with a maximum capacity of 100 material units. The system sets two preset thresholds: normal load when the material quantity is less than 70, warning load when the material quantity is between 70 (inclusive) and 95 (exclusive), and full load lockout when the material quantity reaches or exceeds 95.

[0060] Specifically, in the initial stage of production, the quantity of material in the buffer zone is 20, which is considered a normal load condition. The edge computing device sends a rated speed operation command to the driver of the upstream material conveying mechanism via the industrial control bus, causing the material to enter the buffer zone at a speed of 10 units per minute.

[0061] As production progresses, the amount of material in the buffer area gradually increases. When the amount of material reaches 75 units, the system determines that a warning load condition has been triggered. The edge computing device immediately sends a low-speed operation command to the driver of the upstream material conveying mechanism via the industrial control bus, reducing the material conveying speed to a preset low-speed operation level, such as 5 units per minute. This measure aims to slow down the material inflow, buy time for downstream processing, and prevent rapid material accumulation.

[0062] If the downstream processing speed fails to increase in time and the quantity of material continues to rise, reaching 96 units, the system determines it to be in a full-load lockout state. The edge computing device quickly sends a power-off command to the driver of the upstream material conveying mechanism via the industrial control bus, disconnecting the power supply to the upstream material conveying mechanism and causing it to stop operating completely, thereby stopping material from entering the buffer area. This emergency measure effectively prevents material overflow or blockage, ensuring the safety of equipment and personnel.

[0063] When the material in the buffer zone is consumed by downstream processing and the quantity drops back to 60, the system again determines that it is in a normal operating state. The edge computing device automatically releases the control restrictions on the upstream material conveying mechanism, restores its power supply, and issues a rated speed operation command to restore normal material conveying, thereby ensuring the continuity and efficiency of production.

[0064] The aforementioned load-bearing judgment machine learning model is trained based on historical multi-source sensor data, which includes material quantity, discharge rate, and upstream and downstream cycle time data under three operating conditions: normal operation, early warning operation, and full-load lockout. The load-bearing judgment machine learning model is deployed in an embedded manner on the hardware computing module of the edge computing device.

[0065] The determination of the load-bearing status level of the buffer area includes:

[0066] The edge computing device inputs the correction value of the material quantity and the real-time operating status of the upstream material conveying mechanism into the load-bearing judgment machine learning model to obtain the load-bearing status level;

[0067] When the output of the bearing judgment machine learning model is inconsistent with the threshold-based rule judgment result, the output of the bearing judgment machine learning model is adopted, and the model decision label is output.

[0068] This solution employs a load-bearing judgment machine learning model. Training data covers material quantity, discharge rate, and upstream / downstream cycle time under different operating conditions, including three typical sample types: normal operation, warning operation, and full-load lockout. This allows the model to learn material flow patterns and the dynamic load-bearing characteristics of buffer zones, improving generalization ability. The model is deployed embedded on an edge computing device, allowing for real-time inference close to the data source, avoiding cloud transmission latency and bandwidth bottlenecks, and meeting the needs of rapid decision-making in industrial control. During operation, the model uses the material quantity correction value and the real-time operating status of the upstream conveyor as comprehensive inputs to determine the load-bearing status level. To improve robustness, this solution introduces an intelligent decision-making mechanism: when the model output differs from the preset threshold rule judgment, the model result is prioritized, and a model decision marker is output for monitoring and fault diagnosis. For example, in an abnormal stacking scenario, if the material correction value does not reach the full-load threshold, the rule judgment may misjudge it as "warning load" and issue a low-speed command. However, the model, combining the discharge rate, upstream / downstream cycle time, and other contextual information, can accurately identify that it is actually in "full-load lockout." The system then immediately shuts down and records the decision marker, thereby avoiding overflow and equipment damage risks.

[0069] This application further proposes a more robust material conveying recovery mechanism by introducing a no-load self-inspection and verification process for the docking station to ensure that the buffer area has sufficient material receiving capacity before material conveying is resumed.

[0070] When the correction value of the material quantity in the buffer area falls from the critical difference threshold to below the warning difference threshold due to the receipt of the discharge detection signal, the edge computing device starts the docking station no-load self-inspection and verification process.

[0071] The edge computing device sends a low-speed jog control signal with a duration of a set test time to the frequency converter driver that controls the roller conveyor motor of the connecting platform through the industrial control bus, driving the roller conveyor motor of the connecting platform to run at a preset low-speed jog frequency.

[0072] The edge computing device collects the actual output current sampling value fed back by the frequency converter during the test duration through the analog input port, and calculates the average operating current during the jogging test based on the collected current sampling sequence. The average operating current characterizes the mechanical load resistance borne by the current drive motor of the docking platform.

[0073] The edge computing device compares the average operating current with a preset no-load upper limit current threshold. If the average operating current is greater than the no-load upper limit current threshold, the edge computing device determines that the material discharge detection signal is a false signal, indicating that the docking station has not actually been cleared. It maintains the disconnected state of the power supply circuit of the upstream material conveying mechanism and outputs a jamming alarm signal. If the average operating current is less than or equal to the no-load upper limit current threshold, the edge computing device determines that the docking station has sufficient material receiving capacity. It restores the power supply to the upstream material conveying mechanism through the industrial control bus, allowing the production line to resume its material pushing and unloading operation.

[0074] Specifically, when the correction value of the material quantity in the buffer area falls below the warning difference threshold from the critical difference threshold due to the discharge detection signal (collected by modules such as photoelectric beam, ultrasonic wave, or image recognition), the edge computing device triggers the no-load self-check verification process of the docking station: a low-speed jog control signal (lasting for several seconds, several revolutions per second) is sent to the roller conveyor motor frequency converter driver via the industrial control bus. During the test duration, the output current sampling value fed back by the frequency converter driver is collected in real time through the analog input port, and the average operating current is calculated and compared with the preset no-load upper limit current threshold. If the average current is greater than the threshold, it is determined that the discharge signal is a false signal, the docking station is stuck or not cleared, the upstream power is maintained and a stuck alarm is output; if the average current is less than or equal to the threshold, it is confirmed that the docking station has a material receiving margin, the upstream power supply is restored, and the production line continues to push the material off the line.

[0075] For example, an electronic component docking station performs a no-load self-test.

[0076] The SMT (Surface Mount Technology) assembly workshop of an electronic component manufacturing company features a fully automated assembly line and a 16-story, 24-aisle automated storage and retrieval system (AS / RS). The material handling system includes belt conveyors, roller conveyors, and elevators. The production materials are standardized resistor and capacitor trays, each weighing approximately 0.5 kg and measuring 12 cm long, 8 cm wide, and 2 cm high.

[0077] A buffer zone approximately three meters long and 0.8 meters wide is set up between the belt conveyor and the entrance to the automated warehouse. The buffer zone uses a roller conveyor structure, and the drive motor of the connecting platform has a rated power of 1.5 kilowatts, with speed control via a Delta VFD015E21A frequency converter. The frequency converter supports Profinet communication and can upload actual output current sampling values ​​(analog output range 4-20mA, corresponding to a current range of 0-15A) via the industrial control bus.

[0078] During system operation, an anomaly occurred at the downstream automated warehouse's inbound storage location, triggering a full-load interlock. This caused a power outage on the upstream belt conveyor, stopping material feeding, and the buffer area became full of materials. After maintenance personnel remotely resolved the inbound storage location fault, materials from the buffer area were gradually pushed into the automated warehouse. The photoelectric sensor at the outlet of the transfer station continuously detected material departure events within 30 seconds, and the material quantity correction value gradually decreased from 28 pieces to 18 pieces, triggering the recovery condition judgment logic.

[0079] The specific execution process is as follows: When the edge computing device detects that the quantity of materials has dropped from the critical difference threshold of 28 pieces to below the warning difference threshold of 21 pieces, it initiates the no-load self-check and verification process of the docking station. The edge computing device sends a low-speed jogging control signal with a duration set to 1.5 seconds and a jogging frequency set to 8 Hz to the frequency converter driver via Profinet, corresponding to the docking station drive motor running at a low speed of approximately 0.24 meters per second.

[0080] Upon receiving the jog signal, the frequency converter drive propelled the docking station's drive motor for 1.5 seconds. During this test duration, the analog input port of the edge computing device continuously acquired the actual output current value fed back by the frequency converter drive at a 1-millisecond sampling interval, resulting in a current sampling sequence containing approximately 1,500 sampling points. After moving average filtering, the average operating current was 1.1 amperes.

[0081] The preset no-load upper limit current threshold is 1.2 amps, calculated by multiplying the measured no-load current of the connecting platform drive motor (0.9 amps) by a safety factor of 1.3. Since the average operating current of 1.1 amps is less than the no-load upper limit current threshold of 1.2 amps, the edge computing device determines that the mechanical load resistance borne by the connecting platform drive motor is in a no-load state, there is no material on the connecting platform, the discharge detection signal is real and valid, and the connecting platform has a material receiving margin.

[0082] The edge computing device immediately output a safety relay closing signal via Profinet, restoring the power supply circuit to the upstream belt conveyor driver, and the production line resumed its material feeding and unloading operations. The entire no-load self-check process of the docking station was completed within two seconds, without significantly affecting the production cycle.

[0083] As a comparative verification, on another production line of the same model of electronic components, the no-load self-check mechanism of the docking station was removed, and the recovery logic was triggered solely by the material departure event of the photoelectric through-beam sensor. In a similar abnormal scenario at the receiving location, due to the tilting of the material stack, some material remained at the edge of the docking station. The departure signal from the photoelectric through-beam sensor was actually a false signal, and the system mistakenly judged that the docking station was empty, resuming upstream material pushing. This caused material overflow from the docking station, resulting in a 23-minute production line downtime and the loss of approximately 3,000 units of material. This comparative verification confirms the key value of the no-load self-check mechanism of the docking station based on motor current feedback in actual production.

[0084] This application further proposes a method for detecting the communication status of the data storage server in the detection center. When the communication status is abnormal, the local system prevents the central scheduling commands from driving the upstream material conveying mechanism, and transfers control of the upstream material conveying mechanism to the buffer area based on the load-bearing status level judgment result. This method includes:

[0085] The communication module of the edge computing device is configured with a lightweight write transaction test request to initiate a single-line write transaction request containing a random sequence number to the central data storage server.

[0086] Measure the round-trip processing time of the write transaction, and determine the actual processing load of the core transaction processing engine of the central data storage server based on the round-trip processing time;

[0087] When the actual processing load of the core transaction processing engine meets the preset conditions, the scheduling instruction channel is set to a restricted verification reception state.

[0088] The received scheduling instructions are double-checked with both timestamp and version number to identify and block backlogged or expired instructions.

[0089] Newly generated valid instructions that meet the verification conditions are released under controlled conditions, and the locally maintained physical verification version number is updated.

[0090] Specifically, the communication module of the edge computing device periodically sends lightweight write transaction test requests (single-line random sequence number writing, consuming minimal resources) to the central data storage server. The actual processing load of the core transaction processing engine is evaluated by measuring the round-trip processing time of the request. When the round-trip time exceeds a preset threshold but the connection is not completely severed, the scheduling instruction channel is set to a restricted verification reception state. The received instructions undergo dual verification of timestamp and version number: timestamp verification filters out historical backlog instructions generated during communication blockage, and version number verification compares the warehouse inventory view version number attached to the instruction with the latest physical verification version number maintained locally to prevent invalid instructions generated based on expired data. Only newly generated valid instructions that pass both verifications are allowed to pass under controlled conditions and used to update the local physical verification version number, ensuring that the subsequent verification benchmark remains up-to-date. In a specific implementation, the communication module initiates a test every 500 milliseconds. The normal round-trip time is 50 milliseconds. If the test continues for 200 milliseconds, a restricted verification reception state is triggered. At this time, instructions generated more than 5 minutes ago or with inconsistent version numbers will be blocked. Only instructions with timestamps within the most recent 10-second valid window and version numbers consistent with the local version are allowed to pass. This avoids blindly executing outdated instructions when communication is unstable and ensures effective coordination between local control and central scheduling.

[0091] This application further proposes the following steps for performing dual verification of timestamps and version numbers on received scheduling instructions to identify and prevent backlogged or expired instructions:

[0092] Record the start timestamp of communication blockage and the end timestamp of communication recovery;

[0093] Obtain the instruction generation timestamp, compare the instruction generation timestamp with the start timestamp and the end timestamp respectively, and determine whether the instruction is a historical backlog instruction generated during the communication blockage period;

[0094] Obtain the warehouse inventory view version number of the instruction, compare the warehouse inventory view version number with the locally maintained physical verification version number, and determine whether the instruction is an expired or invalid instruction.

[0095] Based on the determination result of whether the instruction is a historical backlog instruction generated during the communication blockage, and the determination result of whether the instruction is an expired instruction, the execution of the historical backlog instruction or the expired instruction is blocked.

[0096] Allow newly generated valid instructions that meet the verification conditions.

[0097] Specifically, the edge computing device records the start timestamp when communication changes from normal to abnormal and the end timestamp when it recovers, defining the precise period of communication congestion. During the verification phase, each received instruction undergoes a dual check: first, the instruction generation timestamp is compared with the start and end timestamps to identify historical backlogged instructions generated during the congestion period; second, the warehouse inventory view version number attached to the instruction is compared with the locally maintained physical verification version number to identify invalid instructions based on expired inventory views. All instructions determined to be historical backlogged or expired are intercepted and not executed. Only instructions that were neither generated during the congestion period nor based on the latest inventory view are considered newly generated valid instructions and allowed to proceed.

[0098] Taking a manufacturing and warehousing environment as an example, communication between the edge computing device and the central data storage server was interrupted at 10:00 AM and resumed at 10:30 AM. During this period, the central scheduling system generated three instructions: Instruction A (generation time 10:05, version number V1.0), Instruction B (generation time 10:15, version number V1.1), and Instruction C (generation time 10:40, version number V1.2). After communication was restored, the edge computing device recorded the start timestamp 10:00 and the end timestamp 10:30. Instructions A and B, both generated between 10:00 and 10:30, were initially identified as backlogged instructions and intercepted. Furthermore, their version numbers V1.0 and V1.1 were both lower than the local physical verification version number V1.2, further classifying them as expired and invalid instructions. Instruction C, generated at 10:40, is later than its end timestamp of 10:30, and its version number, V1.2, matches the local version number. Meeting the criteria of being neither historically backlogged nor expired, it is allowed to be executed as a newly generated valid instruction. Through dual verification of timestamp and version number, the edge computing device, after communication is restored, only executes new instructions based on the latest data state and generated during the period of normal communication, avoiding potential errors caused by executing outdated instructions.

[0099] Furthermore, this embodiment can be applied to a fully automated packaging line in a pharmaceutical packaging company. The production line is a high-speed filling line, the automated storage and retrieval system (AS / RS) is a cold storage structure, and the material conveying line includes multiple belt conveyors and pneumatic conveying pipelines. The central data storage server is located in the company's data center, running the ERP system and warehouse management system, and is responsible for issuing scheduling instructions, inventory view synchronization instructions, and inbound / outbound scheduling instructions to the edge computing devices in the workshop. The edge computing devices use Advantech MIC-770V3 industrial PCs, which communicate with the upstream belt conveyor drivers and hardware acquisition modules via the Profinet industrial control bus, and with the central data storage server via gigabit Ethernet.

[0100] During a network cutover operation, a firmware upgrade of the enterprise's core switch caused a communication interruption of approximately seven minutes between the workshop edge computing devices and the central data storage server. During this period, the central data storage server, whose scheduling engine continued to operate according to its pre-defined schedule, generated approximately forty scheduling instructions during the communication disruption, which were queued and awaiting issuance once communication was restored.

[0101] The edge computing device detected an anomaly within the first second of the communication interruption. Its communication module initiated lightweight write transaction test requests to the central data storage server every five seconds. Specifically, this involved writing a record containing a millisecond-level timestamp, a random sequence number, and a 16-byte payload to a random primary key row of a dedicated heartbeat table, to probe the actual processing load of the core transaction processing engine. The round-trip time of the first write transaction suddenly increased from the normal 15 milliseconds to over 2000 milliseconds (connection timeout). The write transactions in the subsequent two heartbeat cycles both failed due to connection timeouts. Based on the preset condition that "the round-trip time for three consecutive heartbeat cycles exceeds 100 milliseconds, or any single round-trip time exceeds 500 milliseconds," the edge computing device determined that the actual processing load of the core transaction processing engine of the central data storage server was too high, and the communication status entered an abnormal state.

[0102] When the communication status is determined to be abnormal, the edge computing device immediately sets the scheduling command channel to a restricted verification reception state. During this period, the upstream belt conveyor operates according to the local load status level assessment: when the number of materials in the buffer area is around twelve, it maintains normal load rated speed; when the number of materials rises to twenty-three, it drops to low speed under warning load; when the number of materials exceeds twenty-eight, it enters full load lockout and power-off shutdown. Throughout the entire communication anomaly period, buffer area management is entirely decided locally by the edge computing device and is not affected by central scheduling commands.

[0103] Seven minutes later, the core switch firmware upgrade was completed, and network communication was restored. The central data storage server, the instant communication was restored, pushed all approximately forty scheduling instructions that had been backlogged during the communication blockage to the edge computing device. The edge computing device performed dual verification of the timestamp and version number for each instruction, identifying and blocking historically backlogged or expired instructions.

[0104] The specific verification process is as follows. The edge computing device records the start timestamp T_start of the communication blockage (corresponding to the network cutover start timestamp) and the end timestamp T_end of the communication recovery (corresponding to the return timestamp of the first successful heartbeat after network connectivity). The device obtains the instruction generation timestamp T_gen for each scheduled instruction. If T_gen is between T_start and T_end, the instruction is determined to be a backlogged instruction generated during the communication blockage, and its execution is directly blocked. Simultaneously, the instruction view version number V_instr, which is the version number of the inventory view snapshot at the time the central data storage server generated the instruction, is obtained. The physical verification version number V_local, maintained locally by the edge computing device, records the inventory view version number on which the last successfully executed instruction was based. In this embodiment, V_local was last updated to 182 before the communication blockage.

[0105] Of the backlog of instructions in this batch, thirty instructions had a V_instr of 178 (generated before communication was blocked), which was less than V_local. These were identified as expired and invalid instructions based on the expired inventory view and were directly blocked from execution. The other ten instructions, although their T_gen was after T_start, had a V_instr of 183 (newly generated after communication was restored). They also passed the timestamp and version number verification. The edge computing device controlled the release and execution of these ten newly generated valid instructions, and updated V_local to 183.

[0106] The entire dual verification process was completed within 0.5 seconds of communication restoration, effectively preventing the execution of thirty backlogged instructions based on expired views and allowing only ten newly generated valid instructions to proceed. As a comparative verification, on the same production line without the dual verification mechanism, the thirty backlogged instructions were executed all at once, including multiple forced material push instructions to the buffer area. These directly overwrote the "full load lockout power-off instruction" just issued locally by the edge computing device, causing material overflow in the buffer area, a 23-minute production line shutdown, and the loss of approximately 3,000 units of material. This embodiment verifies the key security value of the scheduling instruction filtering mechanism based on timestamp and version number dual verification in communication anomaly scenarios.

[0107] A multi-source data fusion algorithm is deployed and runs on the hardware computing module of the edge computing device to fuse multi-source sensor data, identify material entry and exit events, and determine the initial quantity of material in the buffer area, including:

[0108] The hardware acquisition module includes a photoelectric beam sensor and an ultrasonic distance sensor, and uploads the acquired data through the industrial control bus.

[0109] The system receives the pulse signal output by the photoelectric sensor, starts a timer, and simultaneously acquires the distance value sequence output by the ultrasonic distance sensor.

[0110] Based on the pulse signal and the distance value sequence, a material entry / exit event is determined; wherein, when the distance value sequence is continuously less than a preset distance threshold, it is determined as a valid material entry / exit event;

[0111] Based on the material entry and exit events, the quantity of materials in the buffer area is added or subtracted.

[0112] When the pulse signal is present but the distance value sequence does not change continuously within the material passage time set by the timer, it is determined to be environmental interference or a false signal, and the addition or subtraction of the material quantity is not performed.

[0113] When the distance value sequence changes but the pulse signal is absent, it is determined that the material stacking height is abnormal or the sensor is in a blind spot. The material quantity is not added or subtracted, and an abnormal alarm signal is output.

[0114] When the difference between the quantity of materials and the previous valid count exceeds the preset allowable deviation range, the edge computing device triggers the image recognition module to perform visual detection on the buffer area, and the quantity of materials is corrected based on the result of the visual detection.

[0115] The hardware acquisition module of this application integrates a photoelectric through-beam sensor and an ultrasonic distance sensor. The two work together to provide multi-source sensing data: the photoelectric through-beam sensor outputs a pulse signal when material passes by to detect the passing event, and the ultrasonic distance sensor outputs a distance value sequence to measure the stacking height of the material. Both types of data are uploaded to the edge computing device in real time via an industrial control bus.

[0116] Specifically, when the edge computing device receives a pulse signal from the photoelectric sensor, it starts an internal timer and simultaneously collects the distance value sequence from the ultrasonic distance sensor. If the distance value sequence is continuously less than a preset threshold within the material passage time set by the timer, a valid material entry / exit event is confirmed, and the initial quantity of material in the buffer area is added or subtracted in real time. The system introduces bidirectional anomaly detection to improve robustness—if there is only a pulse signal and no continuous change in distance value, it is judged as environmental interference or a false signal and is not counted; if there is only a change in distance value and no pulse signal, it is judged as an abnormal material stacking height or a sensor blind zone, and is not counted and an anomaly alarm is output. When the difference between the fused count and the previous valid result exceeds the preset allowable deviation, the edge computing device triggers the image recognition module to perform visual inspection of the buffer area and corrects the initial quantity based on the image analysis results.

[0117] This solution combines two sensors based on different principles—photoelectric through-beam sensors and ultrasonic distance sensors—to achieve accurate identification of material entry and exit events and initial determination of material quantity.

[0118] Specifically, when the difference between the material quantity and the previous valid count result exceeds the preset allowable deviation range, the image recognition module is triggered by the edge computing device to perform visual detection on the buffer area. Based on the visual detection result, the material quantity is corrected. This process includes:

[0119] A material recognition machine learning model trained based on historical material images is constructed and deployed, and the material recognition machine learning model is deployed in an embedded manner on the hardware computing module of the edge computing device;

[0120] An image recognition module containing a visible light camera and a polarizing light camera is deployed above the buffer area;

[0121] The visible light camera acquires a conventional image of the buffer area; the polarized light camera acquires a polarized light image of the buffer area.

[0122] The conventional image is enhanced to improve its contrast and brightness;

[0123] The polarization of the polarized light image is calculated to obtain the reflection characteristics of the material surface;

[0124] An enhanced image of the material is generated by fusing the conventional image and the reflective properties.

[0125] The enhanced image is input into the material recognition machine learning model to obtain the material outline and region segmentation results;

[0126] The quantity of the material is counted based on the material outline and the region segmentation result;

[0127] The quantity of material in the buffer area is corrected based on the count of the material quantity.

[0128] When there is a systematic deviation between the recognition result of the material recognition machine learning model and the actual physical quantity, an adaptive calibration process is initiated: image data and actual physical quantity of the current material are collected; the recognition deviation of the material recognition machine learning model is calculated based on the image data and the actual physical quantity; the recognition parameters of the material recognition machine learning model are adjusted based on the recognition deviation; the material image data and actual physical quantity are collected repeatedly, and the recognition deviation is calculated until the recognition deviation meets the preset calibration accuracy requirements.

[0129] Specifically, the material recognition machine learning model, trained on a large number of historical material images, is deployed embedded on an edge computing device, leveraging the low latency of the edge to achieve rapid localized recognition. The image recognition module consists of a visible light camera and a polarized light camera: the visible light camera acquires conventional visual information, while the polarized light camera captures surface polarization information to eliminate reflections, enhance texture, and distinguish materials; the conventional image undergoes enhancement processing such as histogram equalization or gamma correction to improve contrast and brightness, while the polarized light image extracts reflection characteristics such as polarization degree and polarization angle through polarization calculation. The two are fused to generate an enhanced image containing more comprehensive information, which serves as the model input to output accurate contours and region segmentation results, enabling reliable counting under stacked or partially occluded conditions. When the model's counting results show a persistent and systematic deviation from the actual physical quantity, the system automatically initiates an adaptive calibration process: acquiring new image data and actual physical quantities, calculating the recognition deviation, and automatically fine-tuning the model's recognition parameters accordingly, repeating the iteration until the deviation meets the preset calibration accuracy. This ensures long-term high-precision counting capability even with changes in material batches or environmental fluctuations, without manual intervention.

[0130] Specifically, the above-mentioned method enhances the image input material recognition machine learning model to obtain material contours and region segmentation results, including the following steps:

[0131] Image preprocessing is performed on the enhanced image to improve image quality; adaptive thresholding is applied to the preprocessed image to initially distinguish materials from the background; connected component analysis is performed on the initial segmentation results to identify each connected component in the image and calculate the geometric properties of each connected component; the connected components are filtered according to a preset range of material geometric properties to identify candidate regions that meet the material characteristics; watershed segmentation based on distance transform is performed on the candidate regions to separate tightly stacked materials; post-processing is performed on the watershed segmented regions to optimize the segmentation results; the post-processing includes: smoothing the boundaries of each watershed segmented region to eliminate burrs and irregular edges; calculating the area and perimeter of each smoothed region; initially filtering the regions according to a preset range of material size and shape factor thresholds; merging regions with an area smaller than a preset minimum material area threshold into the adjacent material region with the largest area; merging two regions when the boundary contact length of adjacent regions exceeds a preset adhesion length threshold and the sum of the shape factors of the two regions is less than a preset merge shape factor threshold; and smoothing the boundaries of all material regions.

[0132] Specifically, image preprocessing includes noise reduction, contrast enhancement, and brightness adjustment, aiming to eliminate noise and enhance effective information, providing high-quality input for subsequent segmentation. Adaptive thresholding dynamically adjusts the threshold based on the characteristics of local image regions, which can more accurately distinguish between material pixels and background pixels than global thresholding. Connected component analysis scans the binarized image, identifies all interconnected pixel regions, and calculates their geometric properties such as area, perimeter, and centroid; a preset range of material geometric properties is used to filter connected regions that meet the material characteristics, excluding noise or non-material regions.

[0133] Distance transform-based watershed segmentation treats the image as a topographic map, simulating the process of water flowing from low to high altitudes to segment the image into different catchment basins, thereby effectively separating mutually adhered materials. Post-processing further optimizes the watershed segmentation results: boundary smoothing eliminates jagged or irregular parts on the segmentation boundaries; area and perimeter calculations are used to screen and verify material regions; small areas are merged into adjacent large areas to solve the oversegmentation problem; and adjacent areas are merged when they meet specific adhesion conditions to solve the undersegmentation problem.

[0134] This solution employs a series of refined steps, including image preprocessing, adaptive thresholding, connected component analysis, distance-transform-based watershed segmentation, and post-processing, to work together to achieve high-quality material contours and region segmentation results even when materials are densely stacked, unevenly illuminated, or noisy.

[0135] refer to Figure 3 , Figure 3 This is a schematic diagram of a management optimization system based on artificial intelligence and big data algorithms provided in an embodiment of the present invention. It is applied in a manufacturing and warehousing environment, which includes a production line, an automated storage and retrieval system (AS / RS), and a material conveying line. The system includes:

[0136] The edge computing device is connected to the hardware acquisition module, the driver of the upstream material conveying mechanism, and the central data storage server in the buffer area via an industrial control bus. The edge computing device includes a hardware computing power module and a communication module.

[0137] The hardware acquisition module includes a photoelectric beam sensor, an ultrasonic distance sensor, and an image recognition module, which are used to acquire multi-source sensor data and visual detection images of materials entering and leaving the buffer area, and upload them to the edge computing device through the industrial control bus.

[0138] The signal acquisition unit, deployed on the hardware computing module of the edge computing device, is used to fuse the multi-source sensor data, identify material entry and exit events, and determine the initial quantity of materials in the buffer area. When the difference between the initial quantity and the previous valid count result exceeds the preset allowable deviation range, the image recognition module is triggered to perform visual detection. Based on the result of the visual detection, the initial quantity is corrected to obtain a corrected value for the quantity of materials.

[0139] The load-bearing judgment unit is deployed on the hardware computing module of the edge computing device. It is used to run a load-bearing judgment machine learning model trained based on historical multi-source sensor data. Based on the comparison relationship between the correction value of the material quantity and multiple preset thresholds, it determines the load-bearing status level of the buffer area. Based on the load-bearing status level, it sends a graded adjustment command to the driver of the upstream material conveying mechanism through the industrial control bus. The graded adjustment includes at least reducing the speed of material entering the buffer area or stopping the material from entering the buffer area.

[0140] The control release unit, deployed on the hardware computing module of the edge computing device, is used to automatically release the control restriction on the upstream material conveying mechanism and restore material conveying when the correction value of the material quantity in the buffer area meets the preset recovery conditions.

[0141] The communication detection unit, deployed on the hardware computing module of the edge computing device, is used to detect the communication status of the central data storage server through the communication module. When the communication status is abnormal, it locally blocks the central scheduling command from driving the upstream material conveying mechanism and transfers the control of the upstream material conveying mechanism to the execution of the bearing status level judgment result.

[0142] The management optimization system proposed in this application integrates multiple functional units on an edge computing device in the manufacturing and warehousing environment, enabling real-time, autonomous management of material buffer areas. This system effectively addresses communication latency and data inconsistency issues that may arise in central data storage servers under high concurrency access pressure, thus avoiding the drawbacks of traditional centralized management solutions such as material accumulation and production stoppages. Through edge-side data acquisition, fusion, correction, intelligent judgment, and hierarchical control, this system significantly improves the operational resilience, efficiency, and security of the manufacturing and warehousing environment.

[0143] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A management optimization method based on artificial intelligence big data algorithms, characterized in that, The method is executed by an edge computing device deployed in a manufacturing and warehousing environment, which includes production lines, automated storage and retrieval systems (AS / RS), material conveying lines, and buffer zones. The edge computing device communicates with hardware acquisition modules within the buffer zones, drivers of upstream material conveying mechanisms, and a central data storage server via an industrial control bus. The method includes: The photoelectric through-beam sensor and ultrasonic distance sensor in the hardware acquisition module collect multi-source sensor data of materials entering and leaving the buffer area, and upload the multi-source sensor data to the edge computing device through the industrial control bus. On the hardware computing module of the edge computing device, a multi-source data fusion algorithm is deployed to perform fusion processing on the multi-source sensor data, identify material entry and exit events, and determine the initial quantity of materials in the buffer area. When the difference between the initial quantity and the previous valid count result exceeds the preset allowable deviation range, the edge computing device triggers the image recognition module to perform visual detection on the buffer area. Based on the result of the visual detection, the initial quantity is corrected to obtain the corrected value of the material quantity. On the hardware computing module of the edge computing device, a load-bearing judgment machine learning model trained based on historical multi-source sensor data is run. Based on the comparison relationship between the correction value of the material quantity and multiple preset thresholds, the load-bearing status level of the buffer area is determined. According to the load-bearing status level, a graded adjustment command is sent to the driver of the upstream material conveying mechanism through the industrial control bus. The graded adjustment includes at least reducing the speed of material entering the buffer area or stopping the material from entering the buffer area. When the correction value of the material quantity in the buffer area meets the preset recovery conditions, the control restriction on the upstream material conveying mechanism is automatically released by the edge computing device, and material conveying is restored. The communication module of the edge computing device detects the communication status of the central data storage server. When the communication status is abnormal, the device locally blocks the central scheduling command from driving the upstream material conveying mechanism and transfers control of the upstream material conveying mechanism to the load status level judgment result.

2. The management optimization method based on artificial intelligence big data algorithms according to claim 1, characterized in that, The load-bearing status levels include at least normal load-bearing, early warning load-bearing, and full load lockout; the multiple preset thresholds correspond one-to-one with the load-bearing status levels. The step of issuing graded adjustment commands to the driver of the upstream material conveying mechanism via the industrial control bus based on the load condition level includes: When the load-bearing state is normal load-bearing, the edge computing device sends a rated speed running command to the driver of the upstream material conveying mechanism through the industrial control bus; When the load condition is a warning load, the edge computing device sends a low-speed operation command to the driver of the upstream material conveying mechanism through the industrial control bus to reduce the operating speed to a preset low-speed operation level. When the load condition is full load lockout, the edge computing device sends a power-off command to the driver of the upstream material conveying mechanism through the industrial control bus to disconnect the power supply to the upstream material conveying mechanism and stop the material from entering the buffer area.

3. The management optimization method based on artificial intelligence big data algorithms according to claim 1, characterized in that, The load-bearing judgment machine learning model is trained based on historical multi-source sensor data, which includes material quantity, discharge rate, and upstream and downstream cycle time data under three operating conditions: normal operation, early warning operation, and full-load lockout. The load-bearing judgment machine learning model is deployed in an embedded manner on the hardware computing module of the edge computing device. The determination of the load-bearing status level of the buffer area includes: The edge computing device inputs the correction value of the material quantity and the real-time operating status of the upstream material conveying mechanism into the load-bearing judgment machine learning model to obtain the load-bearing status level; When the output of the bearing judgment machine learning model is inconsistent with the threshold-based rule judgment result, the output of the bearing judgment machine learning model is adopted, and the model decision label is output.

4. The management optimization method based on artificial intelligence big data algorithms according to claim 1, characterized in that, When the correction value of the material quantity within the buffer area meets the preset recovery conditions, the control restriction on the upstream material conveying mechanism is automatically released, and material conveying is resumed, including: When the correction value of the material quantity in the buffer area falls from the critical difference threshold to below the warning difference threshold due to the receipt of the discharge detection signal, the edge computing device starts the docking station no-load self-inspection and verification process. The edge computing device sends a low-speed jog control signal with a duration of a set test time to the frequency converter driver that controls the roller conveyor motor of the connecting platform through the industrial control bus, driving the roller conveyor motor of the connecting platform to run at a preset low-speed jog frequency. The edge computing device collects the actual output current sampling value fed back by the frequency converter during the test duration through the analog input port, and calculates the average operating current during the jogging test based on the collected current sampling sequence. The average operating current characterizes the mechanical load resistance borne by the current drive motor of the docking platform. The edge computing device compares the average operating current with a preset no-load upper limit current threshold. If the average operating current is greater than the no-load upper limit current threshold, the edge computing device determines that the material discharge detection signal is a false signal, indicating that the docking station has not actually been cleared. It maintains the disconnected state of the power supply circuit of the upstream material conveying mechanism and outputs a jamming alarm signal. If the average operating current is less than or equal to the no-load upper limit current threshold, the edge computing device determines that the docking station has sufficient material receiving capacity. It restores the power supply to the upstream material conveying mechanism through the industrial control bus, allowing the production line to resume its material pushing and unloading operation.

5. The management optimization method based on artificial intelligence big data algorithms according to claim 1, characterized in that, The communication status of the central data storage server is detected. When the communication status is abnormal, the central scheduling command is locally blocked from driving the upstream material conveying mechanism, and the control of the upstream material conveying mechanism is transferred to the load-bearing status level judgment result of the buffer area, including: The communication module of the edge computing device is configured with a lightweight write transaction test request to initiate a single-line write transaction request containing a random sequence number to the central data storage server. Measure the round-trip processing time of write transactions, and determine the actual processing load of the core transaction processing engine of the central data storage server based on the round-trip processing time; When the actual processing load of the core transaction processing engine meets the preset conditions, the scheduling instruction channel is set to a restricted verification reception state. The received scheduling instructions are double-checked with both timestamp and version number to identify and block backlogged or expired instructions. Newly generated valid instructions that meet the verification conditions are released under controlled conditions, and the locally maintained physical verification version number is updated.

6. The management optimization method based on artificial intelligence big data algorithms according to claim 5, characterized in that, The process of performing dual verification of the timestamp and version number on the received scheduling instructions to identify and prevent backlogged or expired instructions includes: Record the start timestamp of communication blockage and the end timestamp of communication recovery; Obtain the instruction generation timestamp, compare the instruction generation timestamp with the start timestamp and the end timestamp respectively, and determine whether the instruction is a historical backlog instruction generated during the communication blockage period; Obtain the warehouse inventory view version number of the instruction, compare the warehouse inventory view version number with the locally maintained physical verification version number, and determine whether the instruction is an expired or invalid instruction. Based on the determination result of whether the instruction is a historical backlog instruction generated during the communication blockage, and the determination result of whether the instruction is an expired instruction, the execution of the historical backlog instruction or the expired instruction is blocked. Allow newly generated valid instructions that meet the verification conditions.

7. The management optimization method based on artificial intelligence big data algorithms according to claim 1, characterized in that, The multi-source data fusion algorithm deployed on the hardware computing module of the edge computing device is run to fuse the multi-source sensor data, identify material entry and exit events, and determine the initial quantity of material in the buffer area, including: The hardware acquisition module includes a photoelectric beam sensor and an ultrasonic distance sensor, and uploads the acquired data through the industrial control bus. The system receives the pulse signal output by the photoelectric sensor, starts a timer, and simultaneously acquires the distance value sequence output by the ultrasonic distance sensor. Based on the pulse signal and the distance value sequence, a material entry / exit event is determined; wherein, when the distance value sequence is continuously less than a preset distance threshold, it is determined as a valid material entry / exit event; Based on the material entry and exit events, the quantity of materials in the buffer area is added or subtracted. When the pulse signal is present but the distance value sequence does not change continuously within the material passage time set by the timer, it is determined to be environmental interference or a false signal, and the addition or subtraction of the material quantity is not performed. When the distance value sequence changes but the pulse signal is absent, it is determined that the material stacking height is abnormal or the sensor is in a blind spot. The material quantity is not added or subtracted, and an abnormal alarm signal is output. When the difference between the quantity of materials and the previous valid count exceeds the preset allowable deviation range, the edge computing device triggers the image recognition module to perform visual detection on the buffer area, and the quantity of materials is corrected based on the result of the visual detection.

8. The management optimization method based on artificial intelligence big data algorithms according to claim 7, characterized in that, When the difference between the material quantity and the previous valid count result exceeds a preset allowable deviation range, the edge computing device triggers the image recognition module to perform visual detection on the buffer area, and corrects the material quantity based on the visual detection result, including: A material recognition machine learning model trained based on historical material images is constructed and deployed, wherein the material recognition machine learning model is deployed in an embedded manner on the hardware computing module of the edge computing device; An image recognition module containing a visible light camera and a polarizing light camera is deployed above the buffer area; The visible light camera acquires a conventional image of the buffer region, and the polarized light camera acquires a polarized light image of the buffer region. The conventional image is enhanced to improve its contrast and brightness; The polarization of the polarized light image is calculated to obtain the reflection characteristics of the material surface; An enhanced image of the material is generated by fusing the conventional image and the reflective properties. The enhanced image is input into the material recognition machine learning model to obtain material contours and region segmentation results; The quantity of materials is counted based on the material outline and the region segmentation result; The quantity of material in the buffer area is corrected based on the count of the material quantity. When there is a systematic deviation between the recognition result of the material recognition machine learning model and the actual physical quantity, an adaptive calibration process is initiated: image data and actual physical quantity of the current material are collected; based on the image data and actual physical quantity, the recognition deviation of the material recognition machine learning model is calculated; based on the recognition deviation, the recognition parameters of the material recognition machine learning model are adjusted; image data and actual physical quantity of the material are collected repeatedly, and the recognition deviation is calculated until the recognition deviation meets the preset calibration accuracy requirements.

9. The management optimization method based on artificial intelligence big data algorithms according to claim 8, characterized in that, The step of inputting the enhanced image into the material recognition machine learning model to obtain material contours and region segmentation results includes: Image preprocessing is performed on the enhanced image to improve image quality; Adaptive thresholding is applied to the preprocessed image to initially distinguish between materials and background; Connectivity analysis is performed on the preliminary segmentation results to identify each connected region in the image and calculate the geometric properties of each connected region. Based on a preset range of material geometric properties, the connected regions are filtered to identify candidate regions that match the material characteristics; The candidate region is subjected to watershed segmentation based on distance transformation to separate closely stacked materials; The regions segmented by the watershed are post-processed to optimize the segmentation results. The post-processing includes: smoothing the boundaries of each watershed-segmented region to eliminate burrs and irregular edges; calculating the area and perimeter of each smoothed region; performing preliminary screening of the regions based on preset material size ranges and shape factor thresholds; merging regions with areas smaller than a preset minimum material area threshold into adjacent material regions with the largest area; merging two regions when the boundary contact length of adjacent regions exceeds a preset adhesion length threshold and the sum of the shape factors of the adjacent regions is less than a preset merging shape factor threshold; and smoothing the boundaries of all material regions.

10. A management optimization system based on artificial intelligence big data algorithms, applied in a manufacturing and warehousing environment, wherein the manufacturing and warehousing environment includes production and processing lines, automated storage and retrieval systems, material conveying lines, and buffer areas, characterized in that, The system includes: The edge computing device is connected to the hardware acquisition module, the driver of the upstream material conveying mechanism, and the central data storage server in the buffer area via an industrial control bus. The edge computing device includes a hardware computing power module and a communication module. The hardware acquisition module includes a photoelectric beam sensor, an ultrasonic distance sensor, and an image recognition module, which are used to acquire multi-source sensor data and visual detection images of materials entering and leaving the buffer area, and upload them to the edge computing device through the industrial control bus. The signal acquisition unit, deployed on the hardware computing module of the edge computing device, is used to fuse the multi-source sensor data, identify material entry and exit events, and determine the initial quantity of materials in the buffer area. When the difference between the initial quantity and the previous valid count result exceeds the preset allowable deviation range, the image recognition module is triggered to perform visual detection. Based on the result of the visual detection, the initial quantity is corrected to obtain a corrected value for the quantity of materials. The load-bearing judgment unit is deployed on the hardware computing module of the edge computing device. It is used to run a load-bearing judgment machine learning model trained based on historical multi-source sensor data. Based on the comparison relationship between the correction value of the material quantity and multiple preset thresholds, it determines the load-bearing status level of the buffer area. Based on the load-bearing status level, it sends a graded adjustment command to the driver of the upstream material conveying mechanism through the industrial control bus. The graded adjustment includes at least reducing the speed of material entering the buffer area or stopping the material from entering the buffer area. The control release unit, deployed on the hardware computing module of the edge computing device, is used to automatically release the control restriction on the upstream material conveying mechanism and restore material conveying when the correction value of the material quantity in the buffer area meets the preset recovery conditions. The communication detection unit, deployed on the hardware computing module of the edge computing device, is used to detect the communication status of the central data storage server through the communication module. When the communication status is abnormal, it locally blocks the central scheduling command from driving the upstream material conveying mechanism and transfers the control of the upstream material conveying mechanism to the bearing status level judgment result.

Images